Application Development in Engineering Optimization with Matlab and External Solvers Aalto University School of Engineering

Contens The aim of the lecture is to courage you to use programming, optimization and computational analyses to speed up the design processes Motivation Exercise Contents –Flow chart of optimization –Pre-processing –Analysis –Post-processing Literature 1.Romanoff, J., “Optimization of web-core steel sandwich decks at concept design stage using envelope surface for stress assessment”, Engineering Structures, Vol. 66, 2014, pp User manuals: Matlab, Abaqus, Ansys, Etc

Motivation Analysis of large complex structure involves lots of work –Changes due to prototyping and optimization –Numerous analyses are needed (vibration, flows, ultimate strength) –Numerous documents on analyses to be provided to authorities in form of reports –Etc Some of these tasks can be automatized  requires programming This can be done is all stages of analyses, i.e. pre- and post processing as well as analysis itself Most of the solvers have their own programming language –Abaqus, Python –FEMAP, API/VB –Etc.

Exercise Create a simple script or pseudo code that would automatically create a input file to an external solver and reads the result from output file. Show how you would perform looping for example in terms of optimization. The external solver can be for example: –FE-code: Abaqus (Python), FEMAP (API/VB) –Some executable file –Etc. Report –The idea of the code –The steps the application performs –How well it works and what would be the natural way to extend

Motivation The structures are becoming more advanced and optimized –Lightweight –Sustainable –Safe Effectiveness often requires minimization or maximization of property(ies) of the structure under given load cases and constraints Optimization is mathematical method to find optimal solution –We need optimization algorithms for search of the optimum –We need constraints to make the design feasible in practice –The key issue is to balance both constraint assessment and optimization algorithms – cost vs. accuracy The key question is what to optimize (dimensions, materials, shape, topology), under which conditions (loads, variable range, rules) and for what objective (mass, cost, safety, all of these)

Motivation hchc twtw 2p t f1 t f2 5 variable problem 15 different design possibilities for every variable 1 s for evaluating structural response

Flow of Optimization There are several algorithms available –Matlab central –Internet –Commercial codes, e.g. modeFRONTIER Often you need to combine several software to run different types of analyses –Flow solution –Heat transfer –FEA –Etc Some of these analyses take time and you need to be able to control the process We go through an example containing each of these parts and touch the things you need to pay attention to

Relatively new method – proposed in 1995 Concept based on bird or fish swarm behavior and how knowedlge is tranferred Belongs to the group of evolutionary algorithms – similar principles as genetic algorithm Particle swarm optimisation (PSO) Identify the conceptual design alternatives Population-based optimisation technique developed by Kennedy and Eberhart (1995) Best particle in current calculation round redirects particles of next round to previous best particle One-way sharing mechanism, which looks only for the best solution only All particles tend to converge to the best solution Example of Optimization Algorithm Particle Swarm Optimization (PSO)

Relatively new method – proposed in 1995 Concept based on bird or fish swarm behavior and how knowedlge is tranferred Belongs to the group of evolutionary algorithms – similar principles as genetic algorithm Identify the conceptual design alternatives Population-based optimisation technique developed by Kennedy and Eberhart (1995) Best particle in current calculation round redirects particles of next round to previous best particle One-way sharing mechanism, which looks only for the best solution only All particles tend to converge to the best solution PSO

Speed of particle ‘ i’ at iteration ‘ k’ in design space - Particle’s best location until iteration ‘ k’ - Swarm’s best location until iteration ‘ k’ - Weight factors for the three direction components - Exploration of the design space

Pre-processing The large complex structure can be built automatically using parametric modeling The logic is –Definition of a strake variables –Definition of strake lines and key points –Extrusion of a strake –Making connecting lines between strake end points –Defining areas based on lines –Meshing the areas –Assemble all strakes

Pre-processing Ansys

Pre-processing Programming can be also used to change the input file properties –Equivalent stiffness of shells (e.g. in scantling optimization) –Offset beams and their properties –Nodal coordinates in geometrical optimization –Etc Pay attention to input file format –Space –Tab –Enter Process 1.Create FE mesh 2.Calculate the equivalent stiffness, e.g. in Matlab 3.Print the result in right format to input file Abaqus input file

Pre-processing Matlab Creates Input for Abaqus

Analysis The analysis can be controlled by programming, e.g. in optimization sequence of tasks is important –Create new design –Analyze the design with FEM for Static Dynamic –Extract the FE-results to optimization algorithm Another example is that during the analysis user defined material model can be used. This is coded in FE solver programming language Call input files with numerical info Run the structural analysis with all tasks

Analysis Assign the variables Combine the input files Create equivalent stiffness properties

Analysis Run the vibration analysis in Abaqus and wait that it is Completed (identified by.sta &.lck file existence/non-existence) Collect the results

Post-Processing Some tasks in post processing can be automated –Critical stress check –Critical displacement check –Lowest eigenfrequency check –Etc –Reporting Create result file for Matlab Open Abaqus result file Pick the value of interest Close result file for Matlab

Conclusions Often we need to run analyses several times during optimization using various external solvers This requires automatization, file handling, timing of processes etc Matlab is good environment for of controlling such processes: –Contains many open source optimization algorithms –Can handle external solvers in batch-mode –Has good visualization options –Can create executable with GUI